JPH06187489A - Character recognizing device - Google Patents

Abstract

PURPOSE:To exactly extract an area even in the case of a document for which the character intervals of binary data fetched from a scanner are wide. CONSTITUTION:After image data fetched from the scanner are reduced by an image data reduction part 8, a position where black picture elements are linked is detected by a circumscribed rectangle extraction part 9, and the information of the circumscribed rectangle of linked black picture elements is stored in a memory. When a character rectangle identification part 10 identifies the rectangle of character candidates based on the size of the circumscribed rectangle, the number of picture elements and the density of black picture elements, a character interval/line interval detection part 11 detects the character interval/ line interval both from an inter-rectangle distance between the rectangles adjacent to the up, own, right and left sides of the character candidate rectangle and from the number of times of appearance. On the condition of (character interval < line interval < area interval), a line merging part 12 extracts a line by the merging processing of rectangles with the valid merging distance of character candidate rectangles as the line interval. Then, a character area merging part 13 extracts the character area based on the extracted line, and a character recognition part 14 recognizes characters.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention captures a document image by using an optical means such as a scanner in order to make a database of a printed document and reuse the document, and then, a character, a graphic, The present invention relates to a character recognition device that extracts a region for each attribute such as a table and performs a recognition process according to each attribute.

[0002]

2. Description of the Related Art A conventional character recognition device checks the connection state of black pixels of binary data captured by a scanner, and coordinates of a rectangle circumscribing a block in which black pixels are connected (hereinafter referred to as "circumscribing rectangle"). By detecting the, storing in memory, extracting a circumscribing rectangle of a size equal to or smaller than a predetermined threshold value as a character candidate, and integrating the character candidate rectangles (hereinafter referred to as “character candidate rectangles”). , To extract a row. Then, with respect to the reference character candidate rectangle (hereinafter referred to as "reference rectangle"), the closest character candidate rectangle (hereinafter referred to as "shortest distance rectangle") in four directions of up, down, left and right is detected, and as shown in FIG. For documents with narrow character areas,
To prevent line integration across character areas as shown in Fig. 10, if the distance between the reference rectangle and the detected character candidate rectangle is less than half the rectangle size of the reference rectangle, the two rectangles are within the same character area. Rectangles are merged, assuming that they exist in. At this time, if the shortest distance rectangle detected as the reference rectangle is located vertically, it is determined to be a vertically-assembled character string, and if it is located to the left or right, it is determined to be a horizontally-assembled character string. Therefore, an area is extracted from the detected row information, and if the widths of adjacent rows in the same direction are equal and the distance between the rows is equal to or less than a threshold value, it is determined that the two rows are in the same area and the areas are integrated. After that, character extraction and recognition processing is performed on the extracted character area.

[0003]

However, in the conventional method, since the effective distance of rectangle integration is set to less than half of the rectangle size of the reference rectangle, the character interval is more than half the character size as shown in FIG. The document has a problem that the area is divided into a plurality of areas.

The present invention has been made in view of the above problems, and an object thereof is to provide a character recognition device capable of accurately extracting a region even in a document having a wide character space.

[0005]

SUMMARY OF THE INVENTION The present invention, in order to reduce the character integration process, a means for reducing image data captured by a scanner and a circumscribed rectangle for the reduced data to detect the size of the circumscribed rectangle. Means for identifying a rectangle having a predetermined threshold value or less as a character candidate rectangle, means for determining the mode size of all identified character candidate rectangles as the reference character size of the recognition target document, and for each character candidate rectangle Detecting adjacent rectangles on the left and right, calculating the distance between the rectangles, and storing the number of appearances and the distance as the distance between the rectangles having the largest number of appearances as the horizontal distance between the rectangles of the recognition target document. The rectangles that are vertically adjacent to the character candidate rectangle are detected, the distance between the rectangles is calculated, and the number of appearances and the distance are defined as the distance between the rectangles with the largest number of appearances, which is the vertical distance between the rectangles of the recognition target document. Memorize Means, from the rectangular distance and the number of occurrences of their rectangular distance and vertical detected horizontal, mojikumi direction, between characters, and means for detecting the line space,
Using the condition (between characters <line spacing <between areas), the rectangle integration processing is performed with the integrated effective distance of the character candidate rectangle as the line spacing.
It is composed of means for extracting a line and means for extracting a character region based on the extracted line and cutting out and recognizing a character for each character region.

[0006]

According to the present invention, the effective distance for integrating rectangles is set to the line spacing, so that the region extraction can be accurately performed even in a document having a wide character space.

By separately performing the line integration processing in the horizontal direction and the vertical direction, respectively, it is possible to deal with a mixed document of vertical writing and horizontal writing.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of an apparatus for executing area division in one embodiment of the present invention. Reference numeral 1 denotes a central processing unit (hereinafter referred to as “CPU”) for area extraction,
As shown in FIG. 2, the CPU 1 includes an image data capturing unit 7, an image data reducing unit 8, a circumscribing rectangle extracting unit 9, a character rectangle identifying unit 10, a character spacing / line spacing detecting unit 11, a line integrating unit 12, and a character. It consists of a region integration unit 13 and a character recognition unit 14. 2 is a read-only memory (hereinafter referred to as “RO
3) Random access memory (hereinafter referred to as “RAM”) for storing the image data read by the scanner 4 and the data of the recognition program. 5 is an external CPU.
Keyboard for giving commands to 1, 6 is CPU
1 is a CRT that displays the recognition result recognized by 1.

The character area extracting process will be described below with reference to the flowcharts of FIGS. 3 and 4 and the image data example of FIG.

The image data capturing section 7 captures the document to be recognized as binary image data by the scanner 4 [step 1 (each step is represented as "S1" below)] and the captured binary image data. In order to perform the character area extraction processing of (1) at high speed in the image data reducing unit 8, for example, OR reduction is performed to a resolution of about 100 DPI and the result is stored in the RAM 3 (S
2).

The circumscribing rectangle extraction unit 9 detects a portion where the black images are connected from the reduced image data, extracts the circumscribing rectangle of the connected black images, and coordinate information of all the circumscribing rectangles. Is stored in the RAM 3 (S3). For example, the image data shown in FIG. 5 is extracted as eight circumscribing rectangles K1 to K8 shown in FIG. In the coordinate system of this circumscribed rectangle, the upper left is the origin, the horizontal coordinate is x, and the vertical coordinate is y, and the circumscribed rectangle is the upper left coordinate (x1, y1) and the lower right coordinate (x2, It is represented by y2).

The character rectangle identifying unit 10 identifies a character or non-character based on the size of the circumscribing rectangle, the number of black pixels, and the density of black pixels, and extracts a character candidate rectangle (S4). In this case, to identify as an attribute rectangle other than a character, any one of the following five conditions may be satisfied. That is, 1) When the number of black pixels of the circumscribed rectangle is the threshold value 2) When the aspect ratio of the rectangle is the threshold value FD_RATIO or higher 3) When the short side of the circumscribed rectangle is the threshold value CHAR_MAX or higher 4) The long side of the circumscribed rectangle is the threshold value CHAR_MAX or higher When the black pixel density of the circumscribed rectangle is less than or equal to the threshold value PER_DIAG_MIN 5) When the long side of the circumscribed rectangle is greater than or equal to the threshold value CHAR_MAX and the black pixel density of the circumscribed rectangle is greater than or equal to the threshold value PER_DIAG_MAX. Identified as a candidate rectangle. For example, the rectangles K1 to K8 in FIG. 6 are all identified as character candidate rectangles.

The character-to-character / line-to-line detecting unit 11 performs the processing of S5 to S12 based on the character candidate rectangle.

First, the width and height of each character candidate rectangle are calculated from the character candidate rectangles extracted by the character rectangle identifying unit 10, and the one having the largest number of appearances is set as the reference character size of the recognition target document. (S5). However, when the character size of the character candidate rectangle is less than the threshold value POINT_06, the reference character size is set to POINT_06, and when there are a plurality of character sizes having the same number of appearances, the reference character size is set to the maximum size. For example, the character size of the character candidate rectangle in FIG. 6 is 5, 8 and 9, and the number of appearances thereof is 4, so the reference character size is set to the largest number of 9.

Next, the most frequent distance between the horizontal rectangles is detected. A rectangle adjacent to each of the character candidate rectangles is detected, the distance is calculated, and the number of appearances of the distance between the rectangles is set as the horizontal inter-rectangle distance of the recognition target document, and the appearance count is determined by the RAM 3 The distances are stored in the hCount of the RAM 3 and in the hDis of the RAM 3 (S5).
For example, the distance hDis between the horizontal rectangles of the circumscribing rectangle in FIG. 6 is 5 and the number of appearances hCount thereof is 3.

Similarly, the mode distance between the vertical rectangles is detected, and the distance between the vertical rectangles is calculated as vDis of the RAM 3.
Then, the number of appearances is stored in vCount of the RAM 3 (S5). For example, the distance vDis between the vertical rectangles of the circumscribing rectangle in FIG. 6 is 8, and the number of appearances vCount thereof is 4.

Character spacing and line spacing are determined from the distance between the horizontal and vertical rectangles thus detected and the number of appearances.
(S6). Number of occurrences of horizontal distance between rectangles hCount
And the number of times vCount of appearance of the distance between the rectangles in the vertical direction are different, the process of S7 is performed, and if there is no difference, the process of S10 is performed. At this time, regarding the difference between hCount and vCount, it is determined that there is a difference if the condition of | hCount−vCount | ÷ (hCount + vCount) × 100> threshold CT_DIFF is satisfied. For example, in FIG. 6, the calculation result is 14%, which is smaller than the threshold value CT_DIFF. Therefore, it is determined that there is no difference in the number of appearances, and the process of S10 is performed.

When there is a difference in the number of appearances, the number of appearances hCount of the distance between rectangles in the horizontal direction and the number of appearances vCount of the distance between rectangles in the vertical direction are compared (S
7) If vCount is large, it is determined that the document to be recognized is a vertically aligned document, and the vertical direction information is set to the rectangular merge direction flag dirFlag of the RAM 3 and the character pitch chPi is set.
The distance vDis between the rectangles in the vertical direction is set to tch, and the distance hDis between the rectangles in the horizontal direction is set to the line pitch lnPitch (S8). If there are many hCounts, it is determined that the document to be recognized is a horizontal layout document, horizontal direction information is set in the rectangle merge direction flag dirFlag, and the horizontal distance hDis between rectangles in the character pitch chPitch.
Is set to the line pitch lnPitch, and the distance vDis between the rectangles in the vertical direction is set (S9).

When there is no difference in the number of appearances, the horizontal inter-rectangle distance hDis and the vertical inter-rectangle distance v
When Dis is compared (S10), and vDis is shorter than hDis, the recognition target document is determined to be a vertically-assembled document, vertical direction information is set in the rectangular merge direction flag dirFlag, and vertical rectangles are set in the character pitch chPitch. Distance vDi
s is set to the line pitch lnPitch, and the horizontal distance hDis between rectangles is set (S11). And h
If Dis is shorter than vDis, it is determined that the document is a horizontal document, and horizontal information is set in the rectangle merge direction flag dirFlag, the character pitch chPitch is the distance hDis between the horizontal rectangles, and the line pitch lnPitch is the vertical interval. The distance vDis of each is set (S12).

For example, in FIG. 6, there is no difference in the number of appearances, and h
Since Dis is 5 and the distance is shorter than 8 of vDis, S
After performing 12 processes, dirFlag = horizontal direction, chP
Itch = 5, ln_Pitch = 8 are set.

The line integration unit 12 performs line integration from the characters and lines detected in this way (S13). If the rectangle merge direction flag dirFlag is vertical, vertical rectangle integration (S14), horizontal direction. In the order of rectangle integration (S15) of
If the Flag is horizontal, horizontal rectangle integration (S16),
Rectangle integration is performed in the order of vertical rectangle integration (S17) to extract rows. At this time, (character spacing <line spacing <region spacing)
From the condition of, the effective distance of line integration is determined by the character / line spacing detection unit 11
It is the line spacing detected in. For example, in FIG. 6, dirFl
Since ag is in the horizontal direction, horizontal rectangle integration processing is performed, and then vertical rectangle integration processing is performed. The effective distance of line integration at that time is lnPitch = 8, and the result of rectangle integration is as shown in FIG. Here, the integration of the horizontal rectangles and the integration of the vertical rectangles are carried out in order to deal with a vertically and horizontally mixed document.

The character area integrating unit 13 extracts a character area from the line integration result (S18). For example, the character area extracted from the line information of FIG. 7 is as shown in FIG.

Then, character recognition processing is performed on the finally extracted character area (S19).

In this embodiment, NOISE_MIN ≤ 2 FD_RATIO ≤ 30 CHAR_MAX ≤ 100 PER_DIAG_MIN ≤ 15 PER_DIAG_MAX ≤ 80 POINT_06 ≤ 8 CT_FIFF ≤ 20.

[0025]

As described above, according to the present invention,
By setting the effective distance for integrating the rectangles to be the line spacing, it is possible to accurately extract the region even in a document having a wide character spacing.

Further, by performing the line integration processing individually in the horizontal direction and the vertical direction, it is possible to deal with a mixed document of vertical writing and horizontal writing.

[Brief description of drawings]

FIG. 1 is a block diagram of an apparatus for performing area division according to an embodiment of the present invention.

FIG. 2 is a block diagram of an embodiment of the present invention.

FIG. 3 is a flowchart of an embodiment of the present invention.

FIG. 4 is a continuation of the flowchart of FIG.

FIG. 5 is an example of image data used to describe an embodiment of the present invention.

FIG. 6 shows a result of extracting a circumscribing rectangle of the image data of FIG. 5 in an embodiment of the present invention.

FIG. 7 shows a result of extracting rows from the circumscribed rectangle of FIG. 6 in an embodiment of the present invention.

FIG. 8 shows a result of extracting regions from the row of FIG. 7 in an embodiment of the present invention.

FIG. 9 shows an example of image data in which character regions are closely spaced.

FIG. 10 shows a result of extracting a circumscribing rectangle of the image data shown in FIG.

[Explanation of symbols]

1 ... Central processing unit (CPU), 2 ... Read only memory (ROM), 3 ... Random access memory (RAM),
4 ... Scanner, 5 ... Keyboard, 6 ... CRT,
7 ... Image data importing unit, 8 ... Image data reducing unit, 9
... circumscribed rectangle extraction unit, 10 ... character rectangle identification unit, 11 ... character / line spacing detection unit, 12 ... line integration unit, 13 ... character area integration unit, 14 ... character recognition unit.

Claims (1)

[Claims] 1. A means for reducing image data of a binarized character recognition target document, a portion where black pixels are connected to each other is detected, and information on a circumscribed rectangle of the connected black pixels is stored in a memory. And a means for identifying a character candidate rectangle from the size of the circumscribing rectangle, the aspect ratio of the circumscribing rectangle, and the black pixel density, and a horizontal rectangular interval and a vertical rectangular interval of the character candidate rectangle. And a unit for detecting a character space and a line space, a unit for extracting the line by integrating the character candidate rectangles based on the line space, and a unit for integrating the line and extracting a character region. A character recognition device characterized in that